3D Scene Change Detection from Satellite Imagery
The use of reliable change detection from remote sensory data coming from Earth Observation (EO) technologies is one of the main challenges in security research. One of the main open issues is the accuracy and timeliness of using automatic change detection methods in comparison to visual interpretation of changes, which is still the method used in several operational applications of remote sensing. While most of the changes are optically visible and an expert user with knowledge of a facility will be able to detect changes, automatic processes are difficult to achieve due to problems of registration and area/angle coverage of image series and the detected changes are often difficult to quantify.
One possibility to cope with the above mentioned problems is the use of 3D dimensional models of the facility. These are normally extracted from stereo imagery but are nonetheless difficult to extract and in most of the cases different algorithms will yield different results especially for man-made structures. This is mainly due to a non optimal angle of acquisition and the use of algorithms that are more targeted to terrain modelling than for surface modelling.
The paper starts by describing the extraction of Digital Surface Models (DSM) from high resolution stereo imagery. It describes the algorithms used and the optimisations introduced to deal with man-made structures. The second part of the paper concentrates on the detection of changes from the different DSMs including the introduction of quality values for robust 3D change detection. Representative examples from real sites will be presented.
Keywords: 3D, DSM, stereo, satellite imagery, earth observation, change detection